Maize Leaf Disease Recognition Algorithm And System Implementation Based On Deep Convolutional Neural Network

Maize is one of the most widely planted agricultural products in China.The annual output and quality of maize have a great impact on the economic income of farmers planting maize.With the change of weather,various kinds of maize diseases often appear in the cultivation process.Only when the diseases are found as soon as possible can the rapid prevention and control be carried out in time.However,the traditional machine learning technology not only needs to manually select and extract the characteristics of disease spots but also needs to spend a certain amount of time and money to identify maize diseases.The contribution of characteristics of different maize diseases to disease recognition is very different,and the result of artificial selection of characteristics often affects the recognition accuracy.And it’s hard to determine which features are best.In particular,if the background is very similar to other leaves,the results of image segmentation of disaffected leaves may be less accurate.Most methods cannot effectively segment the leaf and corresponding lesion images from their background,which will lead to unreliable disease recognition results.So far,due to the complexity of diseased leaf images,the recognition of crop diseases based on diseased leaf lesions is still challenging research.With the continuous innovation of computer vision,the use of intelligent and automatic means to prevent and control maize leaf diseases can effectively reduce the huge losses caused by the diseases to farmers.Convolutional neural network model can automatically learn advanced features from a large number of original complex maize disease images without a lot of image preprocessing,lesion segmentation,and artificial design feature extraction.On the basis of the above,this thesis studied how to improve the accuracy of maize leaf disease recognition algorithm.The following are the main contents and results of this thesis:1.A maize leaf disease recognition algorithm based on RI-Net network was proposed.The algorithm is composed of baseline model,Inception module and Res Net module.In the baseline model,a 3×3 stack of convolutional layers is used to increase the size of the receptive field region in feature extraction,and then Inception module and Res Net module are fused to increase the diversity of features.Finally,the features learned are sent to the classifier for classification.By comparing the experimental results,the improved algorithm has a higher accuracy than the traditional manual feature extraction method,and the fusion of model structure improves the capability of multi-scale mapping of features.The algorithm can effectively identify different types of maize leaf disease,which provides a reference for further research on maize leaf disease recognition and recognition algorithm.2.An algorithm of maize leaf disease recognition based on attention residual network was proposed.The algorithm is mainly composed of residual network and improved channel attention module.First,the residual structure is used to prevent network overfitting,then the channels in the residual structure are grouped,and then the attention weight is added to the groups to further capture the local detail features of the image.Meanwhile,the global average pooling is adopted to reduce computation.Finally,the feature input classifier is used to classify maize leaf diseases.This thesis show that,compared with different recognition algorithms,the algorithm can effectively identify different maize leaf disease images in the real environment,and has a higher recognition accuracy.Meanwhile,it establishes a foundation for the algorithm to identify maize disease in the real maize field.3.A maize leaf disease recognition system based on Web terminal is built.The system is mainly composed of four main modules,including login module,knowledge popularization module,disease recognition module and user management module.The login and registration module is responsible for recording the user’s account information,the knowledge popularization module is responsible for popularizing the types of corn leaf diseases and disease prevention knowledge to the user,the disease recognition module is responsible for identifying different types of diseases,and the user management module is responsible for managing and updating the user’s operation record.The system can recognize the pictures uploaded by users to distinguish different kinds of corn leaf diseases,and show the popular science knowledge of corn leaf diseases to users at the same time.The test shows that the system can run normally.